Concordance of CSF measures of Alzheimer's pathology with amyloid PET status in a preclinical cohort: A comparison of Lumipulse and established immunoassays

Abstract Introduction We assessed the concordance of cerebrospinal fluid (CSF) amyloid beta (Aβ) and tau measured on the fully automated Lumipulse platform with pre‐symptomatic Alzheimer's disease (AD) pathology on amyloid positron emission tomography (PET). Methods In 72 individuals from the Insight 46 study, CSF Aβ40, Aβ42, total tau (t‐tau), and phosphorylated tau at site 181 (p‐tau181) were measured using Lumipulse, INNOTEST, and Meso Scale Discovery (MSD) assays, and inter‐platform Pearson correlations were derived. Logistic regressions and receiver‐operating characteristic analysis generated CSF cut‐points optimizing concordance with 18F‐florbetapir amyloid PET status (n = 63). Results Measurements of CSF Aβ, p‐tau181, and their ratios correlated well across platforms (r 0.84‐.94, P < .0001); those of t‐tau and t‐tau/Aβ42 correlated moderately (r 0.57‐0.79, P < .0001). The best concordance with amyloid PET (100% sensitivity and 94% specificity) was afforded by cut‐points of 0.110 for Lumipulse Aβ42/Aβ40, 0.087 for MSD Aβ42/Aβ40, and 25.3 for Lumipulse Aβ42/p‐tau181. Discussion The Lumipulse platform provides comparable sensitivity and specificity to established CSF immunoassays in identifying pre‐symptomatic AD pathology.

A decrease in CSF concentration of soluble Aβ1-42 (Aβ42) peptide is one of the earliest changes in preclinical AD, [5][6][7] likely reflecting the aggregation and deposition of Aβ into plaques in the brain. 8 CSF Aβ42/Aβ40 ratio has consistently shown better diagnostic value for AD than Aβ42 alone, 9 perhaps compensating for individual differences in the total production of Aβ and CSF turnover. 10 The Aβ42/Aβ40 ratio has also been found to mitigate the adsorption-related effects of low sample storage volume (less than 1 mL) on measurements of Aβ concentrations by different platforms. 11,12 CSF Aβ42 has a high concordance of 89% to 92% with amyloid PET 13,14 ; this is further improved when using CSF Aβ42/Aβ40 ratio (94% to 98%). 14 Both reduced CSF Aβ42/Aβ40 ratio 15 and increased uptake of amyloid PET tracers including 18 F-florbetapir 16 have been shown to correlate with neuropathologically-confirmed cerebral Aβ deposition.
Multiple analytical platforms are used for measuring core CSF AD biomarkers; for example, INNOTEST (Fujirebio) provides clinically validated enzyme-linked immunosorbent assays (ELISAs) for Aβ42, totaltau (t-tau), and phosphorylated-tau at site 181 (p-tau181). The Meso Scale Discovery (MSD) Aβ triplex electrochemiluminescence assay simultaneously measures Aβ38, Aβ40, and Aβ42. However, despite efforts to standardize biomarker measurements between multiple platforms and laboratories, 17 differences in absolute values between platforms and coefficients of variation remain high, hampering the development of universal cut points for use in clinical settings. Therefore, there is a drive toward validating fully automated platforms that reduce manual steps as a source for variation. One of these automated platforms is the Lumipulse G system (Fujirebio), on which chemiluminescent immunoassays for Aβ40, Aβ42, t-tau, and p-tau181 have been developed, using the same antibodies as the INNOTEST ELISAs.
Recent studies directly comparing measurements by Lumipulse with INNOTEST ELISAs have shown good concordance between the two platforms but reduced intra-and inter-assay variability on the Lumipulse. [18][19][20][21][22] However, systematic differences in absolute CSF Aβ42 concentrations between Lumipulse and INNOTEST platforms have been observed, 18,19 with one study reporting 27% lower concentrations measured by INNOTEST compared to Lumipulse. 18 When assessing the diagnostic accuracy of Lumipulse CSF Aβ and tau in classifying individuals with clinical AD from cognitively normal controls, Lumipulse ratios of Aβ42/Aβ40, Aβ42/t-tau, and Aβ42/p-tau181 were found to have a higher diagnostic accuracy than individual markers. 20,21 Other studies assessed the ability of Lumipulse assays to differentiate clinical AD from non-AD neurological conditions. No significant difference in diagnostic accuracy has been shown between Lumipulse and INNOTEST assays 19 ; again, compared to using individual biomarkers, the Lumipulse Aβ42/t-tau ratio, 19  Lumipulse assays, showed improved agreement with visual amyloid PET when using Aβ42/Aβ40 or Aβ42/t-tau ratios (concordance 93% to 95%), compared with their respective Aβ42 assays (97% to 89%), but the classic INNOTEST Aβ42 assay gave a concordance of 92%. Spiked Aβ40 over a concentration range of 1 to 40 ng/mL led to progressive decrease in values of Aβ42 measured by the classic INNOTEST (with 60% reduction at the highest spiked concentration) and the MSD platform (with 20% at the highest spiked concentration), but not by the modified INNOTEST. 13 Taken together, these results suggest that the classic INNOTEST assay exhibits some non-specificity to Aβ42 measurement due to quenching of signal by Aβ40 levels.
When assessing CSF by Lumipulse in a mixed-memory clinic cohort, Alcolea et al also found a higher concordance with amyloid PET when using the CSF Aβ42/Aβ40 ratio (86%) than when using individual markers (76% to 84%). 24 Kaplow et al used Lumipulse CSF Aβ42 and t-tau cut points to predict amyloid PET status in multiple cohorts and reported the best performance in all cohorts when using the ttau/Aβ42 ratio (concordance 85% to 95%). 25 As yet, no single study has directly compared all four CSF Aβ and tau markers and ratios measured by the Lumipulse platform with established immunoassays and compared platforms according to concordance with amyloid PET in a preclinical setting. In the present study we extend the comparison of individual Lumipulse CSF Aβ40, Aβ42, t-tau, and p-tau181 markers to also include ratios, with direct comparison to the INNOTEST and MSD platforms. We supplement existing knowledge about the possible contribution of Aβ40 interference to differences in measurements of Aβ42 by evaluating all three platforms, and assess concordance of individual markers and ratios with amyloid PET imaging in a preclinical cohort.

2.3
Imaging procedures

TA B L E 1 Immunoassay platforms used and respective measured biomarkers
The ticks show the biomarkers measured by each platform.

CSF assays
For each of the four analytes of interest, CSF measurements were undertaken using the Lumipulse platform and at least one other established immunoassay platform that uses manual steps in the measurement protocol (Table 1).
For measuring Aβ peptides, three assay platforms were used: INNOTEST with low and high values of the analyte(s) of interest were used. Intrarun variation for the run validation controls and inter-run variation using the control CSF samples are shown in Supplementary Table S1.
Measurements were performed according to the manufacturers' instructions.

Aβ40 interference
Investigation of Aβ40 interference with Aβ42 measurements is detailed in the Supplementary Methods.

Statistical analysis
All analyses used Stata v14.2 (Stata Corporation, Texas, USA). Because individual biomarkers have a positively skewed distribution, logtransformation was undertaken before assessing Pearson correlations between individual biomarker values across platforms. Such transformation was not required before assessing correlations between ratios.
All individuals with available CSF data were included in correlation analyses.
Spearman correlation was used to assess the impact of spiking increasing concentrations of Aβ40 on measurements of Aβ42. Significant Aβ40 concentration-dependent interference was shown if the correlation coefficient (rho) between measured Aβ42 and spiked Aβ40 concentration was significantly less than zero.
In the group with full CSF and amyloid PET data, differences in demographic characteristics between amyloid PET -positive and PET-negative groups were assessed using t tests for age at LP, and   (Table S2).

Participant characteristics
Of the individuals included in the analyses against amyloid PET, 13 (20.6%) were PET positive.

Concordance of CSF biomarkers with amyloid PET
The performance of the three platforms in classifying amyloid PET-negative/positive status is shown in Table 3, for those individual biomarkers and ratios that performed better than chance. Concordance of CSF biomarker ratios with amyloid PET SUVR as a continuous variable is shown in Figure 2. The percentage of discordant individuals was low (4% to 11%) and all discordantly classified individuals were CSF-positive and PET-negative, except when the Lumipulse A 42/t-tau ratio was used (one individual was classified as CSF-negative but PET-positive). All discordantly classified individuals were male, and 57% to 67% were APOE 4 carriers. Despite this, incor-porating age, sex, and APOE 4 carrier status as covariates into predictive models did not significantly change the percentage of discordantly classified individuals or type of discordance (Supplementary Table 2 and Supplementary Figure 3).

DISCUSSION
In this study we build on previous validations of Lumipulse measurements of CSF Aβ and tau biomarkers against two other established CSF assay platforms. We report good correlations of measurements of individual biomarkers of CSF Aβ40, Aβ42, t-tau, and p-tau181 between platforms, in agreement with other studies. [18][19][20] We found a stronger correlation between Lumipulse and MSD measurements of Aβ42/Aβ40 ratio compared to Aβ42 alone. The INNOTEST and MSD platforms showed interference by spiked Aβ40 in measurements of Aβ42, but the Lumipulse platform did not. All ratios incorporating Aβ42 were more concordant with amyloid PET than individual biomarkers; the The area under the receiver-operating characteristic curve (AUC), its 95% confidence interval, the Youden index (at which the combination of sensitivity and specificity is maximized), and the corresponding optimal cut point are shown for each of CSF Aβ42, t-tau, p-tau181, and their ratios in predicting amyloid PET status (n = 63). a Higher than AUC for Lumipulse t-tau, De Long test P = .005. b Lower than AUC for Lumipulse p-tau181, De Long test P = .024. Measurements of p-tau181 correlated better than t-tau between Lumipulse and INNOTEST, and this was also reflected in the Aβ42/p-tau181 and Aβ42/t-tau ratios. Our findings for t-tau contrast with high correlations (r >0.9) reported in other studies between Lumipulse and INNOTEST measurements 19,29 but cannot be explained in our study by any differences in pre-analytical handling, as both t-tau and p-tau181 were measured on both platforms from the same aliquot of CSF from each individual. Values of t-tau at the lower end of the range of samples measured were less well correlated, whereas values of p-tau181 were very well correlated throughout the range measured ( Figure 1E,F). It is unclear as to whether this reflects altered performance of one or both of the t-tau assays in this part of the measurement range, but it is important to note that the values were well above the published lower limits of quantification for both assays.
We found differences in absolute biomarker values between platforms, as reported by others. 18,19 For MSD values this could in part be due to different antibodies used for Aβ measurements; other reasons could include differences in the technology and calibrators used. None of the three Aβ42 assays had been calibrated against the CSF Aβ42 certified reference material (CRM), 30 which was developed to provide an international standard for this analyte, and CRMs are not yet available for the other analytes. Furthermore, it is possible that native Aβ40 leads to inaccurate Aβ42 estimates in the INNOTEST assay as reported previously. 13 Lumipulse is less susceptible to matrix effects, based on the optimal minimal required dilution of sample in the conjugate solution. We did find significant A 40 interference with MSD measurements; these could be due to similar matrix effects as found on the INNOTEST, or due to differences in antibody specificity between the two assays.
We report 100% sensitivity across all three platforms for CSF A 42 in predicting cortical amyloid load, but the specificity of the Lumipulse measurements (74%) was lower than that of the INNOTEST (86%) or MSD (80%). In contrast to other studies, which show CSF t-tau and p-tau181 to be good individual predictors of amyloid PET, 24 CRMs, and participants in the other studies having a wider range of cognitive performance and overall higher prevalence of APOE 4 carriage than the participants of our cohort.
Advantages of the Lumipulse platform over conventional assays include reduction in labor-intensive steps and manual error, reduced total analysis time due to testing all four biomarkers on the same CSF sample, and improved accuracy for analyte detection, due to the measurement by photon-counting of direct light emitted rather than wavelength-based colorimetric absorbance. Furthermore, the Lumipulse platform can process small numbers of CSF samples, without needing to collect enough samples to use in batched assays (as is required for the INNOTEST or MSD). However, a disadvantage of the Lumipulse is its requirement for a large dead volume of 100 μL, relative This study has some limitations that might be assessed in future research. MSD analysis took place on a separate day using a separate sample aliquot to that used for INNOTEST and Lumipulse analysis.
Lumipulse measurements were performed in singleton due to CSF volume requirements, so precision of the Lumipulse assays was not assessed. We did not compare measurements of all four biomarkers on all platforms, and we focused on comparing Lumipulse measurement with other immunoassays but not with other methods of measurement like mass spectrometry. Although the AUC obtained for prediction of amyloid PET status were higher for CSF Aβ42 and its ratios than that obtained by the model using age, sex, and APOE 4 carrier status, the differences did not reach statistical significance, likely due to this being an interim data set of samples collected by this point of the ongoing study. This cohort consists mostly of cognitively healthy individuals of the same age. It is possible that some individuals classified as "CSFpositive" (through the use of the ratio cut points) but "PET-negative" do actually have sub-threshold cerebral amyloid deposition, as CSF changes may precede PET changes. 32 However, in the absence of neuropathological data to date in this cohort, the use of amyloid PET as an in vivo "gold standard" is a necessary limitation.
In summary, this study supports the use of the fully automated Lumipulse platform, particularly for measuring CSF Aβ42/Aβ40 and Aβ42/p-tau181, to identify cerebral amyloid deposition with excellent sensitivity and high specificity, without Aβ40 interference, even in cognitively normal individuals.

CONFLICTS OF INTEREST
Avid Radiopharmaceuticals, a wholly owned subsidiary of Eli Lilly, kindly provided the 18 F-florbetapir tracer (Amyvid) free of cost but had no role in the design, conduct, analysis, or reporting of Insight 46 study findings. We are particularly indebted to the support of the late Chris Clark of Avid Radiopharmaceuticals who championed this study from its outset.
Fujirebio provided and set up the Lumipulse platform free of cost but had no role in the design, conduct, analysis, or reporting of this study.